The Science of Prediction

Bemused by a Swiss attempt to analyse a joke, Russ Swan finds real humour in a dubious attempt at the science of prediction 

One of the great joys of reporting on science is the opportunity to be nosey about what goes on in some of the great laboratories and research projects of the world. Those, and some of the less great ones too.

Did you know, for example, that researchers at Imperial College have put in a bid to analyse the purported vitamin D deficiency of goblins? Father and son team Dr Nicholas and Joseph Hopkinson reckon the reason Tolkien's baddies are so bad is because they don't get out enough, while those nice hobbitses benefit from healthy amounts of sunlight. They observe, while apparently keeping a straight face, that "no previous studies have investigated vitamin D levels in imaginary populations."

No kidding! And there was me thinking that Imperial was one of the world's leading universities. I can’t help wondering if they are having a laugh, and if so whether they wouldn’t be better directing their efforts to the famously funny nation of Switzerland.

An electron got pulled over by a traffic cop. "Do you realise you were travelling at 93mph?" demands the policeman. "Oh, great, now I'm lost," replies the electron.

You either get it, or you don't, and it's either funny or it isn't, right? Well that's not good enough for the Swiss National Science Foundation, which is embarking on a new research programme to analyse humour. Laughing at a joke is the result of a two-stage process in the brain, first detecting an incongruity before then resolving it, we're told. This results in a 'feeling of cheerfulness' which can be clearly differentiated from that of other positive emotions, SNSF hilariously continues.

The announcement then reminds us that: "science has demonstrated that animals are capable of planning into the future". This is better, making us anticipate that the researchers are about to claim the detection of a sense of humour in non-humans. The disappointment follows immediately, however: "Fortunately, we can still claim humour as our unique selling point".

You see what they did there? They created an incongruity, and then resolved it! Oh how we laughed. Those Swiss and their wacky sense of playfulness. That, and their pointless raising of an investigative dead-end.

For me though, the highlight of this particular stroll through the research announcements that regularly cross my desk was the brave prediction of a publishing decision. It was made by an international team of computational linguists at Radboud University, Nijmegen, in the Netherlands, and is particularly appealing because it combines the risky business of scientific forecasting with a new pseudoscientific-sounding discipline called digital humanities. As if that was even a real thing.

In December, the team got very excited waiting for the announcement from New York of Time magazine's Person of the Year for 2013. They were excited, because they'd already applied their computational and linguistic skills to predict who would be named the person the magazine's editors felt had "most influenced the news last year, for good or bad".

They did this by the fairly simple expedient of counting up how often various people had been named in the weekly magazine during the year. "It turns out to be very significant", they say, "if a person is continuously mentioned".

So far, this looks like a moderately tedious exercise in counting column inches. It gets more interesting when the researchers confess that they used Wikipedia – that famously accurate and error-free resource – to work out who everybody was. It becomes positively edge-of-the-seat stuff when they further confess that previous analyses have yielded a 29% success rate.

I mean, really? I know humanities is not a proper science, and that media analysis is not exactly a laboratory operation, but what on Earth makes them think that it might be a good idea to make a 'scientific' prediction when they must know it will be wrong seven times out of ten?

You hardly need me to tell you the rest. The Dutch team promised us it would be Barack Obama, and it wasn't. They predicted ten names altogether, none of which made it into the actual top five (the winner was Pope Francis). Hilariously, the team predicted a third place for Miley Cyrus, a singer and twerker, and fifth place for Angelina Jolie, an actress. That's what happens when you base predictions on magazine coverage.

The thing is, I could have told them that Obama wouldn't win it. You see, he was Time's Person of the Year in 2012. You don’t need a PhD in computational linguistics or digital humanities (oh! the humanities!) to work that out. You just needed to read the darned magazine.